GPT-5.6 Sol Max was already the better deal on DeepSWE v1.1 – The recent price cut widens the gap even further. Sol scores 72.7% at $6.47/task, compared with Fable 5 Max at 69.7% and $21.63/task. DeepSWE tests coding agents on 113 original, long-horizon engineering tasks.
ECONOMY
-
DeepSeek Runs 3.6 Times Cheaper in a Matched-Speed Agent Comparison
By
–
cool tool! seeing DeepSeek run 3.6x cheaper than GPT-5.6 Sol with the exact same median time is pretty wild
-
Productivity J-Curve Explains Why Complementary Investments Delay AI Gains
By
–
Yep. As @ChadSyverson
, @danrock and I wrote in 2021, General purpose technologies like AI require significant complementary investments. This delays their effect on productivity and growth. We call this the Productivity J-curve. -
FinanceGym Uses Human Demonstrations for Financial Agent Fine-Tuning
By
–
FinanceGym data can be used for Reinforcement Finetuning or Supervised Finetuning agentic models. It contains human-demonstrated trajectories w/thinking traces, & teaches agents financial work, not just financial knowledge.
-

FinanceGym Releases Over 59,000 Training Samples for Finance Agents
By
–
Training finance agents is hard: Private data is locked inside large corporations. MIT CSAIL researchers release "FinanceGym," the largest finance-agent dataset yet, w/59K+ samples across 37 roles and 40 apps/sites. Even Kimi K3 & Qwen 3.8 Max failed its test set.
-
AI Agent Reviews OpenCV Grant Opportunities and a Rejected Proposal
By
–
What a Consultant Would Have Charged for This
— Satya Mallick (@LearnOpenCV) 24 août 2026
I asked an AI agent to find grants for OpenCV and analyze why our 2023 NSF application was rejected. A consultant would charge $150 to $200/hr and spend a day on it, about $1,600. The agent did a strong first pass for a fraction of… pic.twitter.com/wfMkKsnabhWhat a Consultant Would Have Charged for This I asked an AI agent to find grants for OpenCV and analyze why our 2023 NSF application was rejected. A consultant would charge $150 to $200/hr and spend a day on it, about $1,600. The agent did a strong first pass for a fraction of
-

Enterprise AI Coding Cuts Generation Costs While Review Remains Expensive
By
–
// Applying Anthropic Primitives at Large Enterprises // Frontier models collapsed the cost of writing custom code. But they did not reduce the cost of reviewing and maintaining it. Each bespoke solution drifts from the next. This means that there is a huge tax that goes into
-

Open Models’ Share of Vercel Tokens Reportedly Rises to 62%
By
–


Open-source AI’s token share on Vercel surged from 28% to 62% in just two months. Just today, the Financial Times wrote that Fable 5 usage is declining because it's too expensive and open source is becoming increasingly better. At the same time, I think that more and more users
-
AI Job Shock Debate Distinguishes Hiring Slowdown From Actual Automation
By
–
“We are not experiencing an “AI Jobs Shock”. At most, we are experiencing a reduction in hiring by executives who do not understand the technology but who have absorbed vibes that there will be a real-soon-now “AI Jobs Shock”. Where do these vibes come from? From a combination of
-
AI Data Center Electricity Demand Raises Questions About Clean Power Capacity
By
–
AI has a growing energy problem. Data centres already consume electricity on the scale of major countries, and demand could more than double by 2030. The AI race may ultimately depend not only on chips and models, but also on who can produce abundant, reliable and clean power.
